Beruflich Dokumente
Kultur Dokumente
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
compared with multiple daily injections
both provided with structured education
for adults with type 1 diabetes: a health
economic analysis of the Relative
Effectiveness of Pumps over Structured
Education (REPOSE) randomised
controlled trial
Daniel John Pollard,1 Alan Brennan,1 Simon Dixon,1 Norman Waugh,2
Jackie Elliott,3 Simon Heller,3 Ellen Lee,4 Michael Campbell,1 Hasan Basarir,5
David White,4 On behalf of the REPOSE group
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
People with type 1 diabetes mellitus (T1DM) are unable much of the previous benefit attributed to pumps may be
to produce insulin due to autoimmune destruction of due to the additional education that pumps users require
their insulin-secreting beta cells. Those affected have to when initiating therapy.11 Therefore, uncertainty remains
inject insulin to prevent diabetic ketoacidosis (DKA) in as to the clinical benefit and cost-effectiveness of pumps
the short term, and in the long term, to prevent micro- as a treatment option for adults with T1DM.
vascular and macrovascular disease. Insulin is generally The Relative Effectiveness of Pumps over Structured
administered by intermittent subcutaneous injection with Education (REPOSE) trial was conducted to assess if pumps
the dose adjusted according to eating, physical activity and offered any additional benefit compared with MDI for the
current blood glucose levels. Therapy is designed to keep treatment of adults with T1DM in the UK who are eligible
blood glucose as close to normal as possible, to prevent to attend a structured education course but do not have an
both microvascular complications and to reduce the risk immediate clinical need for a pump. Participants in both
of macrovascular disease.1 A further aim of treatment is trial arms received a DAFNE course, which taught MDI
to achieve as good a quality of life as possible, particularly users to use long-acting and short-acting insulins appropri-
since self-management of the condition is challenging and ately. In this paper, we present the cost-effectiveness analysis
arduous, demanding the implementation of complex skills. of pumps+DAFNE compared with MDI+DAFNE for adults
Historically, insulin was given twice a day, often as in the UK with T1DM who are eligible to receive a struc-
premixed insulin, but such an approach imposes a tured education course but did not have a clinical need for
rigid lifestyle on people with T1DM and makes it diffi- the immediate commencement of pump therapy.
cult to maintain blood glucose levels close to normal.
Most individuals require intensive insulin therapy to
maintain tight glycaemic control. This approach and Methods
its integration within a flexible lifestyle is promoted in Economic evaluation methods
structured training courses, such as the Dose Adjust- The health economic analysis followed the prespecified
ment For Normal Eating (DAFNE)2 course and others.3 health economic analysis plan, which was outlined in the
The principles of this multiple daily injections (MDI) trial protocol paper.12 Two approaches were undertaken
via subcutaneous injection approach involves the use to assess the cost-effectiveness of pumps: an economic
of quick-acting insulin injected before eating (with the evaluation alongside the clinical trial (EEACT) and a long-
dose calculated according to the amount of carbohydrate term economic model to assess the lifetime outcomes.
eaten) combined with long-acting background ‘basal’ In the EEACT, data on the costs and quality-adjusted
insulin, usually given twice daily, to control blood glucose life years (QALYs) were obtained from data collected
in between meals. in the REPOSE trial, whereas the long-term model esti-
As an alternative to MDI, insulin can be administered mated the lifetime costs and QALYs based mainly on
by an insulin pump system in which insulin is delivered the biomedical outcomes collected in the REPOSE trial.
throughout the day using a small, portable pump, connected All modelling analyses took a lifetime time horizon,
by a flexible plastic tube to a subcutaneous cannula. This and all EEACT analyses took a 2-year (trial duration)
technology is relatively expensive compared with MDI. time horizon. The prespecified primary analysis relates
Current National Institute for Health and Care Excellence to the lifetime modelling, and as such, are reported in
(NICE) guidelines recommend that all adults with T1DM this paper. The methods and results of the EEACT are
in the UK receive a structured education course of proven reported in online supplementary material A. In line with
benefit at a clinically appropriate time.4 Insulin pumps are NICE guidance, all economic analyses took a National
a treatment option for adults with T1DM who either have a Health Service and personal social services perspective,
HbA1c above 69 mmol/mol (8.5%) or experience disabling and costs and QALY outcomes were discounted at 3.5%.13
hypoglycaemia.5 Insulin pumps do not replace the need for
education, and it is currently recommended that specialist The economic model
teams should provide structured education programmes The Sheffield Type 1 Diabetes Policy Model version 1.3.2,
and advice on diet, lifestyle and exercise for people using henceforth ‘the model’, is an individual level simulation
an insulin pump.5 model used to estimate the lifetime costs and QALYs asso-
Use of pumps in T1DM varies substantially between ciated with pump+DAFNE and MDI+DAFNE. The most
countries. In England, 11.7% of people with T1DM are recently published version of the model14 was updated
estimated to use a pump, which compares to around 40% using data collected in the REPOSE trial at 12-month
in the USA.6 7 It has been proposed that insulin pumps are and 24-month follow-up. Specifically, the level of HbA1c,
underused in the UK and that the glycaemic control of risk of severe hypoglycaemia, risk of DKA and the prob-
adults with T1DM could be improved if pumps were used ability that an individual would switch insulin delivery
more widely.8 Two NICE appraisals found that there was mechanism and costs related to the intervention (insulin
insufficient evidence on insulin pump therapy compared and MDI consumables, diabetes related contacts with
with MDI in which adults with T1DM used both long- healthcare professionals, insulin pumps and the asso-
acting and short-acting insulins.9 10 Furthermore, a recent ciated consumables) were updated. Relevant literature
also informed the changes in HbA1c beyond the 2-year The simulation cohort
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
trial duration and the probability of death from end-stage To obtain the simulation cohort, participants’ REPOSE
renal disease (see online supplementary material B, page data (n=260) were sampled with replacement 5000
2) in both arms. The model examines disease progression times.17 This created a simulation cohort of 5000 individ-
over a lifetime using an annual time cycle. An individual’s uals for whom some would have missing baseline charac-
HbA1c determines their risk of progression for all diabetic teristics. Each individual from REPOSE was included in
complications in the model, which include: nephropathy, the simulation cohort a median of 19 times (IQR 16–22).
neuropathy, retinopathy, macular oedema, myocardial To obtain data values for these missing variables in the
infarction, stroke, heart failure, angina and severe hypo- simulation cohort, two imputation procedures were used.
glycaemia and DKA. A higher HbA1c increases the risk of A truncated regression procedure for missing continuous
progression for all complications in the model. Individuals variables (baseline: total cholesterol, high-density lipo-
in the model are at risk of death from the incidence of: protein (HDL) cholesterol, systolic blood pressure and
nephropathy, myocardial infarction, stroke, heart failure, cost of insulin used in the year prior to baseline), in which
angina and all-cause mortality. HbA1c indirectly effects variables were limited to positive values, and a Poisson
mortality in the model, as the probability death does not procedure for missing categorical variables (gender). In
differ by HbA1c; however, the risk of experiencing these the truncated regression imputation procedure, all char-
events is higher for someone with a higher HbA1c. The acteristics with complete data in the REPOSE trial popu-
model attaches utilities to health states and costs to lation were used as predictive covariates. In the Poisson
events, allowing the calculation of costs and QALYs over imputation procedure, all characteristics with complete
a lifetime. Full details of how the model calculates the data in the REPOSE trial population and the imputed
incidence of diabetes complications is provided in Heller data for the continuous variables were used as predictive
et al (p. 104–106) and Thokala et al.14 15 The incremental covariates. In both imputation procedures, a single impu-
cost-effectiveness ratio (ICER) calculated was compared tation was used. As this was preformed in the simulation
with the £20 000 to £30 000 per QALY gained threshold cohort rather than the REPOSE data, the replications of
used by NICE.13 In each model run, the life course of the individuals in the simulation cohort had different
5000 individuals was simulated. This number of simu- data values if they had missing baseline data from the
lated individuals was considered to be sufficiently robust trial. Identical individuals were simulated in both model
for decision making (see online supplementary material arms. A summary of the baseline characteristics and
B, page 1). All model analyses were conducted using completeness of data for the 260 REPOSE participants in
SIMUL8 professional 2010.16 the ITT population and the 5000 individuals in the simu-
lation cohort are given in table 1.
The clinical data
The details of the methods used in the REPOSE trial have The clinical effectiveness data
been reported elsewhere.12 The eligibility criteria of the trial Statistical analyses of the REPOSE trial data were
included adults with T1DM if they were eligible to receive a conducted to estimate treatment switching, HbA1c, severe
structured education course, did not have a clinical indica- hypoglycaemia and DKA. The main clinical results paper
tion to receive a pump immediately, as determined by the is available; however, it differs in this paper as it focuses
investigator, or did not have a strong preference to receive a on statistical in those people with a HbA1c greater than
pump.12 The REPOSE trial was conducted in eight centres or equal to 58 mmol/mol (7.5%).18 Information on long-
in England and Scotland, involving 267 individuals. Out term changes in HbA1c was obtained from the available
of these 267 individuals, 260 had HbA1c data for at least literature and the REPOSE trial data. Unless otherwise
one postbaseline follow-up visit (intention to treat (ITT) stated, all statistical analyses were conducted using STATA
population) and 236 adhered to their randomised treat- V.13.1.19 Full details on the statistical methods used, the
ment (per protocol population). Self-reported participant results and how the analyses were incorporated into the
information, which includes the number of DKA events, model are provided in online supplementary material B.
EuroQol EQ-5D-3L, 12 item short form survey (SF-12) and To estimate the incidence of treatment switching in
resource use were collected at baseline, 6 months, 1 year the first and second year of the model, a time-to-event
and 2 years after attending DAFNE training. HbA1c, the analysis was conducted using treatment switching as the
primary endpoint, was measured at baseline and at these event of interest. Kaplan-Meier curves were plotted, and
follow-up appointments. All inpatient admissions, the inci- parametric survival curves were fitted to the Kaplan-Meier
dence of severe hypoglycaemia and whether an individual curves.20 21 Separate parametric survival models were
switched insulin delivery method were collected on an fitted to individuals randomised to pump+DAFNE and
ongoing basis. Information was collected on the staff time MDI+DAFNE. The goodness of fit of the parametric
spent delivering precourse fitting sessions for individuals survival curves was assessed using the Akaike infor-
allocated to receive a pump. The cost of insulin pumps and mation criterion, Bayesian information criterion and
associated consumables in routine practice, and the insulin visual assessment of the fit of the parametric curves to
pump consumables used by the REPOSE participants, were the Kaplan-Meier curves at 1 and 2 years. As treatment
collected from seven of the eight trial centres. switching was included in the model, it is important to
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
Table 1 The baseline characteristics of REPOSE participants and the simulated cohort
Characteristic REPOSE trial population (n=260) Simulated cohort (n=5000)
Continuous variables mean (SD) (% of individuals with data prior to imputation)
Baseline HbA1c (mmol/mol) 76.0 (18.6) (100) 75.9 (18.2) (100)
Baseline HbA1c (%) 9.1 (1.7) (100) 9.1 (1.7) (100)
Age (years) 40.4 (13.4) (100) 40.4 (13.3) (100)
Diabetes duration (years) 18.0 (12.5) (100) 18.0 (12.3) (100)
Triglycerides (mmol/mol) 1.4 (1.0) (100) 1.4 (0.9) (100)
Total cholesterol (mmol/mol) 4.9 (0.9) (99.6) 4.9 (0.9) (99.7)
HDL cholesterol (mmol/mol) 1.6 (0.4) (96.5) 1.6 (0.4) (96.4)
LDL cholesterol (mmol/mol) 2.8 (0.9) (96.2) 2.7 (0.9) (96.1)
Systolic blood pressure 131.4 (16.4) (98.8) 131.3 (16.0) (98.9)
Baseline cost of insulin £357.24 (147.65) (94.8) £360.39 (157.92) (98.4)
Baseline cost of diabetes-related contacts £561.61 (885.92) (100) £571.63 (928.92) (100)
Categorical variables n/N (percentage) (% of individuals prior to imputation)
Gender
Female 104/260 (40.0) (40.0) 1990/5000 (39.8) (39.3)
Male 152/260 (58.5) (58.5) 2950/5000 (59.0) (59.3)
Missing 4/260 (1.5) (1.5) 0/5000 (0.0) (1.4)
Physical activity
Low 67/260 (25.8) (25.8) 1266/5000 (25.3) (25.3)
Medium 128/260 (49.2) (49.2) 2471/5000 (49.4) (49.4)
High 65/260 (25.0) (25.0) 1263/5000 (25.3) (25.3)
Smoking status
Current 50/260 (19.2) (19.2) 951/5000 (19.2) (19.2)
Former 67/260 (25.8) (25.8) 1325/5000 (26.3) (26.3)
Never 143/260 (55.0) (55.0) 2724/5000 (54.5) (54.5)
Race
White 258/260 (99.2) (99.2) 4959/5000 (99.2) (99.2)
Black 2/260 (0.8) (0.8) 41/5000 (0.8) (0.8)
Nephropathy
No complications 239/260 (91.9) (91.9) 4600/5000 (92.0) (92.0)
Microalbuminuria 13/260 (5.0) (5.0) 234/5000 (4.7) (4.7)
Macroalbuminuria 7/260 (2.7) (2.7) 152/5000 (3.0) (3.0)
Dialysis or transplant 1/260 (0.4) (0.4) 14/5000 (0.4) (0.4)
Neuropathy
No complications 238/260 (91.5) (91.5) 4599/5000 (92.0) (92.0)
Neuropathy or ulcers 22/260 (8.5) (8.5) 401/5000 (8.0) (8.0)
Retinopathy
No complications 145/260 (55.8) (55.8) 28/5000 (56.0) (56.0)
Background diabetic retinopathy 91/260 (35.0) (35.0) 1740/5000 (34.8) (34.8)
Proliferative diabetic retinopathy 24/260 (9.2) (9.2) 465/5000 (9.3) (9.3)
Myocardial infarction
No complications 255/260 (98.1) (98.1) 4896/5000 (97.9) (97.9)
Myocardial infarction 5/260 (1.9) (1.9) 104/5000 (2.1) (2.1)
Stroke
No complications 259/260 (99.6) (99.6) 4983/5000 (99.7) (99.7)
Stroke 1/260 (0.4) (0.4) 17/5000 (0.3) (0.3)
Continued
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
Table 1 Continued
Characteristic REPOSE trial population (n=260) Simulated cohort (n=5000)
Heart failure
No complications 259/260 (99.6) (99.6) 4934/5000 (99.6) (99.6)
Heart failure 1/260 (0.4) (0.4) 18/5000 (0.4) (0.4)
Angina
No complications 257/260 (98.9) (98.9) 4934/5000 (98.7) (98.7)
Angina 3/260 (1.2) (1.2) 66/5000 (1.3) (1.3)
HDL, high-density lipoprotein; LDL, low-density lipoprotein; REPOSE, Relative Effectiveness of Pumps over Structured Education.
adjust the modelled treatment effectiveness and cost of initial sample size. An SE could not be directly calculated,
treatment for those individuals who switch. If individuals as we did not have access to the patient level data from the
switched treatment, then their HbA1c was assumed to studies. Instead, data on the SD of the change in HbA1c
change as though they had been allocated to the other between the 1-year and 2-year follow-ups specific to each
arm. No explicit effect of switching on the incidence rate REPOSE trial arm were used to inform the uncertainty in
ratio (IRR) associated with the model arms for either this parameter. These SDs were divided by the square root
severe hypoglycaemia or DKA was included. However, due of the combined sample size of the studies used to inform
to the change in HbA1c associated with switching, individ- the mean effect to estimate SE for this parameter.
uals who switched from pump to MDI were at a lower risk Negative binomial regressions were used to estimate the
of severe hypoglycaemia and at a higher risk of DKA. The risk of severe hypoglycaemia and DKA in the ITT popu-
opposite was true for those individuals who switched from lation of REPOSE using the Zellig package in R V.3.2.0.
MDI to pump. In the base case analysis, the HbA1c effect Negative binomial regressions were fitted separately to
of pump+DAFNE compared with MDI+DAFNE was esti- severe hypoglycaemia and DKA events and to the first and
mated in the per protocol population, as the people who second years of follow-up data. The regressions fitted to
switched treatment were excluded from this population. the second year data were used to estimate the incidence
To estimate each individual’s HbA1c in the model, a of severe hypoglycaemia and DKA in all model time
beta regression was fitted to the HbA1c data collected cycles, except for the first time cycle. This assumption was
in REPOSE for all individuals in the per-protocol popu- based on clinical expert opinion, which was confirmed
lation at the 1-year and 2-year follow-up.22 Bounds were by the trial management group, that the first 6 months of
placed on the beta regression so that HbA1c was between using a pump was associated with a learning period, and
the clinically plausible bounds of 29 mmol/mol (4.8%) after this time, the incidence of severe hypoglycaemia
and 201 mmol/mol (20.5%), which were provided by and DKA would decrease.
two clinical experts in the REPOSE trial management
group. Missing data were observed for HbA1c values in Costs and utilities
the per-protocol population at 6 months (2.1% missing), The costs and health state utility values used for each
1 year (4.2% missing) and 2 years (4.2% missing). Full health state in the model are provided in table 2 and
details on the imputation procedure used to account for table 3, respectively. All costs were reported in 2013/2014
the missing data and the specification of the beta regres- prices. Costs sourced from evidence in previous years
sion are provided in online supplementary material B were inflated to 2013/2014 prices using the hospital
(see p. 8). The effects in a beta regression are not easily and community health services pay and prices index.28
interpretable by themselves; however, information can be The cost of insulin and MDI consumables used, cost of
obtained on the direction of effect. diabetes related contacts with healthcare professionals
The expert opinion of clinical members involved in and cost of the pump intervention costs were estimated
the literature review of clinical studies for the REPOSE separately for both year 1 and year 2 of the REPOSE
trial was sought to identify studies with long-term data trial. The unit cost of insulin and MDI consumables used
on HbA1c for the type of pumps used in the trial and was obtained from the British National Formulary and
people who used MDI after DAFNE. Three articles on the Health & Social Care Information Centre, the unit cost
long-term trends in HbA1c for pump users23–25 and two of diabetes related contacts were obtained from NHS
articles on the long-term trends in HbA1c for MDI users reference costs and unit cost of an insulin pump (with its
post-DAFNE26 27 were used to estimate yearly changes in associated consumables) was obtained from a survey of
HbA1c for each model arm. The average annual increase the REPOSE trial sites.29–31 Table 2 part A shows the six
in HbA1c was calculated for each study by calculating regression formulae that estimate an individual’s annual
yearly increase in HbA1c between the lowest observed cost for each of these components as a function of: their
HbA1c value and the last follow-up point in the study baseline HbA1c, which treatment they start the year on
where the sample size was greater than one quarter of the (MDI or pump), whether they switch treatment and also
Multiplier for the baseline DRC – – +0.11 +0.03 – – REPOSE trial data in year 1.
cost (<<<NO ENTITY>>>1) (year 1 costs).
REPOSE trial data in year 2.
(ongoing costs).
NHS reference costs.30
Pump costing survey.
Multiplier for the baseline +0.97 +1.04 – – – –
insulin cost (<<<NO
ENTITY>>>2)
Multiplier for the baseline +5.08 +12.81 −21.66 +12.15 – –
HbA1c (DCCT % scale) (<<<NO
ENTITY>>>3)
Receiving pump at the start of −517.91 −527.64 +129.08 +88.99 +2056.11 +2050.99
the year (<<<NO ENTITY>>>4)
Switch from pump to MDI +554.47 +153.35 +280.16 −47.10 −1143.68 −905.03
(<<<NO ENTITY>>>5)
Switch from MDI to pump 0† −353.27 +733.95 +201.93 +804.57 +1134.27
(<<<NO ENTITY>>>6)
Constant (<<<NO ENTITY>>>0) +381.77 +324.53 +415.46 +299.80 0.00 0.00
Adverse events
Hypoglycaemia 187 18.69 Heller et al14 DKA with hospitalisation 1399 140 NHS reference costs38
Nephropathy
Ongoing yearly cost of 36 3.56 BNF39 and McEwan Ongoing yearly cost of 36 3.56 BNF39 and McEwan et al40
microalbuminuria et al40 microalbuminuria ongoing
Ongoing yearly cost of ESRD 24 436 2444 NHS reference Death due to ESRD 0 0 Assumed equal to zero
costs38
Neuropathy
Clinically confirmed 271 27.14 Currie et al41 Clinical neuropathy 271 27.14 Currie et al41
neuropathy
Diabetic foot syndrome 2848 285 NHS Reference PAD with amputation 7221 722 NHS Reference costs38
costs38 (year 1)
Ongoing yearly cost of PAD 439 43.93 McEwan et al40
with amputation
Continued
Retinopathy
Background retinopathy 145 14.47 McEwan et al40 Proliferative retinopathy 661 66.11 McEwan et al40
Macular oedema (year 1) 5710 571.0 NICE,42 BNF43 Macular oedema (year 2) 3416 341.6 NICE,42 BNF43
42 43
Macular oedema (year 3) 2562 256.2 NICE, BNF Macular oedema (ongoing) 277 27.7 NICE,42 BNF43
44
Blindness (year 1) 1584 158 Clarke et al Blindness (ongoing) 519 51.88 Clarke et al44
Cardiovascular
First MI (year 1) 6788 679 Clarke et al44 Second MI (year 1) 6788 679 Clarke et al44
*The variance covariance matrices used to parameterise the multivariate normal distribution are provided in online supplementary material B tables 4, 7 and 12.
†This variable was included in the original regressions; however, the model would converge when this covariate was included, so this parameters is taken to be a zero value in the total cost formula.
–This value was not included as a covariate in the regression formula and is taken to be a zero value in the total cost formula.
The cost for each total cost in part A is calculated using the following formula:
Total cost=β0+β1*individual’s baseline diabetes-related contact cost+β2*individual’s baseline insulin cost+β3*individual’s baseline HbA1c (DCCT % scale)+β4*individual’s treatment at the start of the year (1=pump,
0=MDI)+β5*(1=switched from pump to MDI, 0=did not switch from pump to MDI)+β6*(1=Switched from MDI to pump, 0=did not switch from MDI to pump).
BNF, British National Formulary; DAFNE, Dose Adjustment for Normal Eating; DCCT, Diabetes Control and Complications Trial; DKA, diabetic ketoacidosis; DRC, diabetes related contacts; ESRD, end-stage renal
disease; HF, heart failure; MDI, multiple daily injections; MI, myocardial infarction; NHS, National Health Service; NICE, National Institute for Health and Care Excellence; PAD, peripheral arterial disease; pump, insulin
pumps; REPOSE, Relative Effectiveness of Pumps over Structured Education.
Open Access
7
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
Open Access
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
Table 3 The base case utility parameters
Beta distribution
Health state for event Utility SE Alpha Beta Source
Baseline utility value
Male with type 1 diabetes and no 0.866 0.010 947.79 146.90 Peasgood et al45
complications
Gamma distribution
Disutility SE Alpha Beta Source
Complications or covariates
Female with type 1 diabetes and no 0.0236 0.008 8.70 0.003 Peasgood et al45*
complications
Adverse events‡
Severe hypoglycaemia −0.002 −0.002 1 0.002 Peasgood et al45
Diabetic ketoacidosis −0.0091 −0.01 0.83 0.01 Peasgood et al45*
Nephropathy§
Microalbuminuria 0 Assumption
Microalbuminuria −0.017 0.01 2.89 0.01 Coffey et al46
ESRD −0.078 0.026 9 0.01 Coffey et al46
Neuropathy§ (applied to the history of events)
Clinical neuropathy −0.055 0.01 30.25 0.002 Coffey et al46
Clinically confirmed neuropathy −0.055 0.01 30.25 0.002 Coffey et al46
Diabetic foot syndrome −0.1042 −0.119 0.77 0.14 Peasgood et al45
PAD with amputation −0.1172 −0.055 4.54 0.03 Peasgood et al45*
Retinopathy§ (applied to the history of events)
Background retinopathy −0.0544 −0.023 5.59 0.01 Peasgood et al45
Proliferative retinopathy −0.0288 −0.026 1.23 0.02 Peasgood et al45
Blindness −0.208 0.013 256 0.001 Coffey et al46
Cardiovascular§ (applied to the history of events)
MI (first year) −0.065 0.03 4.69 0.01 Alva et al47
MI (subsequent years) −0.057 0.03 3.61 0.02 Alva et al47
Heart failure −0.101 0.032 9.96 0.010 Alva et al47
Stroke −0.165 0.035 22.22 0.007 Alva et al47
Angina −0.09 0.018 25 0.004 Clarke et al48†
*A parameter value was not available in the author’s preferred statistical model.
†Value is presented in table 5 as ischaemic heart disease.
‡These disutilites are applied transiently to the number of these events in each year.
§These disutilites are applied to the history of ever having had one of these events.
ESRD, end-stage renal disease; MI, myocardial infarction; PAD, peripheral arterial disease.
the participants’ self-reported use of insulin and level of model results (see online supplementary material B,
contact with healthcare professionals prior to recruitment page 1). All statistical models, except for the risks of
to the trial. Details on how these parameters were used to severe hypoglycaemia and DKA, for which the outcomes
estimate the costs in the model are provided in table 2. were directly simulated in R V.3.2.0, were included in
Scenario and subgroup analyses the model using a multivariate normal distribution.32
For the base case, a probabilistic sensitivity anal- The long-term changes in HbA1c were assumed to be
ysis (PSA) was conducted where each parameter was distributed using independent normal distributions for
sampled from its probability distribution, the results each model arm. Deterministic scenario analyses were
were recorded, the process was repeated 500 times and undertaken to assess the robustness of the results. The
the averages from these 500 model runs were reported. key scenario analyses for the model include: uncertainty
This number of PSA runs was considered to be suffi- in the estimates of treatment effectiveness, the timing
cient to allow for robust decision making based on the of HbA1c changes in the model, the utility decrement
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
and insulin pump consumables. Clinical effectiveness data
A two-way deterministic threshold analysis was conducted Treatment switching
to assess the HbA1c reduction and/or annual cost reduc- The analysis of treatment switching in the REPOSE trial
tion necessary to potentially make pumps cost-effective suggested that an exponential curve provided the best fit
in the UK. All threshold analysis runs were conducted to the data for individuals receiving pump+DAFNE and a
deterministically. In this analysis, each individual’s HbA1c Weibull curve provided the best fit for individuals in the
was estimated as though they received MDI. Then those MDI+DAFNE arm.
individuals in the pump arm received a fixed change in
HbA1c. This change in HbA1c was varied at 10 different HbA1c
values between −3.3mmol/mol (−0.3%) and −13.1mmol/ The results of the analysis of HbA1c data suggested that
mol (−1.2%), and annual cost of insulin pumps were also pump+DAFNE compared with MDI+DAFNE was associated
varied between 100% of the observed cost in the REPOSE with statistically insignificant HbA1c reductions in both year
trial and 50% of the observed cost. 1 and in year 2. The analysis of the literature for long-term
Deterministic subgroup analyses were conducted in the trends in HbA1c suggested that weighted average annual
following populations: progression of HbA1c was +0.568 mmol/mol (+0.052%) per
1. baseline HbA1c <69 mmol/mol (8.5%) annum in the pump+DAFNE arm and +0.590 mmol/mol
2. baseline HbA1c ≥69 mmol/mol (8.5%) (+0.054%) per annum in the MDI+DAFNE arm. The esti-
3. baseline HbA1c ≥58 mmol/mol (7.5%) mated standard errors for the long-term trends in HbA1c
4. 69 mmol/mol (8.5%) >baseline HbA1c ≥58 mmol/ were of 0.627 mmol/mol (0.040%) for MDI+DAFNE and
mol (7.5%) 0.627 mmol/mol (0.057%) for pump+DAFNE.
5. 80 mmol/mol (9.5%) >baseline HbA1c ≥69 mmol/ These data produce a trace of HbA1c as in figure 1.
mol (8.5%) The solid lines show the average HbA1c in the model for
6. baseline HbA1c ≥80 mmol/mol (9.5%) each year. The dotted lines show what the HbA1c would
7. all individuals in the per protocol population. have been in each year, if a higher HbA1c did not have
The subgroup analyses were conducted by resampling an indirect effect on the number of deaths in the model
the simulation cohort from the individuals who met each or treatment switching effects. The dashed line shows the
of these criteria in the ITT population. number of people remaining alive in each year and is
Figure 1 The trace of: (A) the mean HbA1c in each year of the model, (B) the mean HbA1c ignoring the effects of death and
treatment switching, (C) the number of people alive in each year of the model in both arms over 30 years. DCCT, Diabetes
Control and Complications Trial; MDI, multiple daily injection.
plotted on the secondary axis. As expected from the base in REPOSE, then the analysis shows that the reduction in
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
case clinical results, the MDI+DAFNE arm has a higher HbA1c (for pump compared with MDI) would need to
HbA1c than the pumps+DAFNE arm, and this effect be 12.0 mmol/mol (1.1%) or more, for pump to have an
slightly increases over time ICER below £20 000 per QALY gained. When the annual
cost is 25% lower, then a HbA1c reduction of more than
Severe hypoglycaemia and DKA 7.7 mmol/mol (0.7%) would be needed to have an ICER
The results of the negative binomial models used to esti- below £20 000 per QALY gained. When the annual cost is
mate the incidence of DKA indicated that pump+DAFNE halved, then a HbA1c reduction of 3.3 mmol/mol (0.3%)
compared with MDI+DAFNE was associated with more would be sufficient to have an ICER below £20 000 per
DKA events in the first year (IRR 1.40, 95% CI 0.55 to QALY gained.
3.58) but fewer events in year 2 (IRR 0.93, 95% CI 0.23 to The result of the subgroup analyses are presented in
3.69). The results of the negative binomial models used figure 2C. Similarly to the scenario analysis, no subgroup
to estimate the incidence of severe hypoglycaemia indi- was identified in which pump+DAFNE would be likely
cated that pump+DAFNE compared with MDI+DAFNE to be considered more cost-effective than MDI+DAFNE.
was associated with more severe hypoglycaemic events in The most cost-effective subgroup was those individuals
year 1 (IRR 1.33, 95% CI 0.49 to 3.65) but fewer events in with a baseline HbA1c greater than or equal to 80 mmol/
the second year (IRR 0.35, 95% CI 0.08 to 1.44). mol (9.5%), with an ICER of £96 394 per QALY gained
versus MDI+DAFNE.
Base case results
Table 4 shows the base case results for the long-term
cost-effectiveness analysis using the PSA. For the Discussion
pump+DAFNE arm, the mean costs of the intervention This is the first study to examine the marginal benefits
are £42 124 discounted over the lifetime horizon, which and cost-effectiveness of insulin delivered using pumps
compares with £19 829 for the MDI+DAFNE arm. This over MDI when both groups have had structured educa-
leads to an increase in intervention costs of £21 295 over a tion, in patients without an immediate clinical need for
person’s lifetime. The QALYs lived, prior to applying the an insulin pump, as recommended by NICE. Our find-
utility decrements associated with diabetic complications, ings show that providing pumps for this wider group of
in the pump+DAFNE arm is 13.91 QALYs compared with adults with T1DM would be unlikely to be considered
14.03 QALYs in the MDI+DAFNE arm, a difference of 0.11 cost-effective in the UK because the estimated ICER was
QALYs. The differential incidence of adverse events in the just over £142 000, substantially higher than the £30 000
pump+DAFNE arm compared with the MDI+DAFNE arm per QALY upper limit often used by UK decision makers.
leads to a lower costs (−£371) and higher QALYs (+0.00) This finding is also valid in all scenario and subgroup
per person. The reduced incidence of diabetic compli- analyses examined. The threshold analysis indicates that
cations in the pump+DAFNE arm leads to lower costs if new insulin pump technology is developed and the
(−£2070) and more QALYs (+0.01) per person. The net costs are similar to current insulin pumps, a trial would
incremental lifetime cost of pump versus MDI is there- have to demonstrate a HbA1c reduction in the region of
fore estimated as £18 853 (95% CI £6175 to £31,645) 11 mmol/mol (1.0%) to 13 mmol/mol (1.2%) compared
per person. The net incremental QALY gain per person with MDI+DAFNE for the new technology to be consid-
is 0.13 (95% CI −0.70 to +0.96) QALYs per person. The ered cost-effective in the UK for the population analysed
ICER associated with pump+DAFNE was £142 195 per in this study. Conversely, if insulin pump prices were
QALY gained. This is considerable above the £30 000 halved, then reductions of only 3 mmol/mol compared
per QALY threshold, which is the higher limit at which with MDI are required.
NICE would usually consider interventions to be cost-ef- The key strengths of this study are that it is based on
fective.13 Figure 2A shows that this result is subject to a a rigorously conducted cluster RCT with economic data
small degree of uncertainty, as in 14.0% of the PSA runs, directly collected during the study. Data completeness
pump+DAFNE would be considered as being cost-effec- for the primary outcome was very high at 95%. The
tive using the lower end of the range used by NICE. study uses an individual level simulation model of type
1 diabetes disease progression over a lifetime horizon.
Scenario and subgroup analyses The study does have limitations in terms of the evidence
The results of 11 scenario analyses are presented used to inform the long-term changes in HbA1c, which
in figure 2B. None of the scenario analyses showed were based on five observational studies with follow-up
pump+DAFNE to be cost-effective versus MDI+DAFNE. ranging from 3.7 to 10 years, rather than trials with long
The lowest ICER found (£31 747 per QALY gained) was follow-up periods. Also, as REPOSE is the first study to
for scenario 3 in which yearly cost of pumps and consum- assess the effectiveness of pumps+DAFNE, the long-term
ables was halved. evidence was based on observational studies of pumps
The results of the two-way price and effectiveness in which the education component is likely to have
threshold analysis are given in table 5. When the annual been different from the DAFNE structured education
pump cost is assumed to be 100% of the prices observed provided in the REPOSE trial. Our PSA has incorporated
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
Table 4 Long-term cost-effectiveness analysis: base case results using probabilistic sensitivity analysis
MDI+DAFNE Pump+DAFNE Incremental
Mean lifetime discounted costs per person
Intervention costs
Insulin and MDI consumables £12 215 £5476 −£6740
Diabetes-related contacts £5023 £6289 £1266
Insulin pumps and pump consumables £2228 £28 967 £26 739
DAFNE course £363 £392 £29
Subtotal intervention costs £19 829 £41 124 £21 295
Adverse event costs
Severe hypoglycaemia £133 £41 −£92
Diabetic ketoacidosis £1161 £882 −£279
Subtotal cost of adverse events £1294 £922 −£371
Long-term complication costs
Nephropathy £40 786 £38 853 −£1933
Neuropathy £1859 £1805 −£53
Retinopathy+macular oedema £6365 £6263 −£102
Myocardial infarction £1838 £1844 £6
Heart failure £607 £609 £2
Stroke £253 £254 £0
Angina £1134 £1143 £8
Total cost of long-term complications £52 841 £50 771 −£2070
Total costs £73 964 £92 817 £18 853
(95% CI £6175 to £31 645)
Mean undiscounted life years per person
Total life years 28.3181 28.7999 0.3790
(95% CI −2.7392 to 3.3403)
Mean discounted QALYs per person
QALYs lived 13.9145 14.0292 0.1147
(excluding decrements due to complications)
QALYs lost due to adverse events
Severe hypoglycaemia −0.0014 −0.0004 0.0009
Diabetic ketoacidosis −0.0075 −0.0057 0.0018
Subtotal QALYs due to adverse events −0.0088 −0.0061 0.0027
QALYs lost due to complications
Nephropathy −0.1853 −0.1792 0.0061
Neuropathy −0.3092 −0.3010 0.0082
Retinopathy+macular oedema −0.3316 −0.3293 0.0022
Myocardial infarction −0.0528 −0.0532 −0.0004
Heart failure −0.0385 −0.0387 −0.0002
Stroke −0.0343 −0.0345 −0.0002
Angina −0.0754 −0.0761 −0.0007
Subtotal QALYs lost due to complications −1.0271 −1.0120 0.0152
Total QALYs 12.8785 13.0111 0.1326
(95% CI −0.7087 to 0.9623)
Summary
Total mean discounted costs per person £80 471 £99 337 £18 853
(95% CI £6175 to £31 645)
Continued
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
Table 4 Continued
MDI+DAFNE Pump+DAFNE Incremental
Total mean undiscounted life years per person 28.3181 28.7999 0.3790
(95% CI −2.7392 to 3.3403)
Total mean discounted QALYs per person 12.8785 13.0111 0.1326
(95% CI −0.7087 to 0.9623)
ICER (£/QALY gained) £142 195
Probability that pump+DAFNE is cost-effective at 14.0%
a threshold of £20 000 per QALY gained
DAFNE, Dose Adjustment for Normal Eating; ICER, incremental cost-effectiveness ratio; MDI, multiple daily injections;
pump, insulin pumps; QALY, quality-adjusted life year.
uncertainty around these estimates. Second, the analysis features. Fourth, the results of the threshold analysis used
of rates of DKA was based on self-reported occurrence to determine what the effectiveness of pumps would need
of DKA in REPOSE (the same method as used in other to be considered cost-effective should be interpreted with
recent trials of DAFNE Heller et al, chapter 8),14 rather some degree of caution for two reasons. First, we used
than corroborated data on inpatient admissions that had a fixed HbA1c effect for every individual, rather than
a smaller number of events and could not be analysed a method that accounts for heterogeneous treatment
statistically. Third, only one pump type was assessed in the responses, and second, we assume switchers from MDI to
trial, and some caution may be needed when considering pump get the fixed HbA1c reduction and switchers from
extrapolating results to modern pumps with autosuspend pump to MDI get the same fixed HbA1c increase. In an
Figure 2 The results of the economic analyses presented on the cost-effectiveness plane for: (A) the base case PSAs, (B)
the results of the scenario analyses and (C) the results of the prespecified subgroup analyses. DCCT, Diabetes Control and
Complications Trial; DKA, diabetic ketoacidosis; ICER, incremental cost-effectiveness ratio; MDI, multiple daily injection; PSA,
probabilistic sensitivity analysis; QALYs, quality-adjusted life years.
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
Table 5 The incremental cost-effectiveness ratio associated with insulin pumps for different HbA1c reductions (for all adults with type 1 diabetes mellitus) and annualised
Red, the incremental cost-effectiveness ratio is above £30,000 per quality adjusted life year gained; Orange, the incremental cost-effectiveness ratio is between £30 000 and £20
Dominates
Dominates
the ones in this analysis would likely lead to the same
£17 610
£15 515
£13 419
£11 323
ICERs. Finally, as no one study could provide sufficient
(1.2%)
£9227
£7131
£5036
£2940
£844
information on the disutility decrements associated with
diabetic events for people with type 1 diabetes, the data
come from disparate sources. This means that the magni-
tude of some events appears to be small compared with
Dominates
Dominates
others. For example, end-stage renal disease has a utility
£18 953
£16 730
£14 507
£12 284
£10 060
(1.1%)
£7837
£5614
£3391
£1167
decrement of −0.078, whereas heart failure has a higher
decrement of −0.101.
A 2015 systematic review on the cost-effectiveness of
000 per quality adjusted life year gained; Green, the incremental cost-effectiveness ratio is less than £20 000 per quality adjusted life year gained.
insulin in various countries identified four studies from
the UK, three of which presented an ICER.34 The base case
£26 258
£23 736
£21 213
£18 691
£16 169
£13 646
£11 124
£8602
£6079
£3557
£1035
£37 712, which indicate that in two out of the three studies
that pumps had an ICER within/below the threshold range
usually used by NICE to determine the cost-effectiveness
3.3 mmol/mol 4.4 mmol/mol 5.5 mmol/mol 6.6 mmol/mol 7.7 mmol/mol 8.7 mmol/mol 9.8 mmol/mol
£8585
£6100
£3615
£1130
£8193
£5370
£2548
£9099
£5951
£2803
£64 493
£58 901
£53 309
£47 717
£42 125
£36 533
£30 941
£25 349
£19 757
£14 164
(0.5%)
randomising them to either continued MDI or pumps. Gayna Babington, Gail Bird, Janet Evans, Tasso Gazis, Nicola Maude, Karen Nunnick,
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
While the current literature does provide some indica- Dawn Spick, Laura Fenn, Carla Gianfrancesco, Valerie Gordon, Linda Greaves, Susan
Hudson, Valerie Naylor, Chloe Nisbet, Carolin Taylor, Karen Towse and Candice
tion of which individuals may benefit from pumps, the Ward contributed at sites to participant recruitment, intervention delivery and data
clinical evidence informing these studies was not limited collection. Cindy Cooper, Gemma Hackney, Diana Papaioannou, Emma Whatley
to those individuals who had first received structured and David White provided central trial management, oversight and monitoring.
Mike Bradburn, Michael Campbell, Munya Dimairo and Ellen Lee contributed to the
education. This is important because the effectiveness of
statistics. Hasan Basarir, Alan Brennan, Simon Dixon and Daniel Pollard contributed
insulin pumps would be expected to differ in this group, to the health economics. Nina Hallowell, Jackie Kirkham, Julia Lawton and David
with this being related to observed and unobserved Rankin designed and undertook the qualitative work. Katharine Barnard led the
patient characteristics. One particular hypothesis that quantitative psychosocial work. Timothy Chater and Kirsty Pemberton provided data
management. Fiona Allsop and Lucy Carr provided central administration. Pamela
we believe is worthwhile investigating is whether pumps
Royle conducted literature searches and exploratory analyses. Gill Thompson
should be used in the UK by those adults with T1DM and Sharon Walker provided central DAFNE support. Pauline Cowling conducted
who actively self-manage after attending a structured the fidelity assessment. Henry Smithson provided user representation on the
education course, but either have not achieved the NICE management group.
target HbA1c levels of less than or equal to 48 mmol/mol Contributors DJP contributed to the design of all of the health economic analyses
(6.5%) or who have problematic hypoglycaemia. Another and conducted all statistical and modelling analyses used in this paper. AB oversaw
the design and implementation of the modelling analyses. SD, oversaw the design
important research question concerns the effectiveness and implementation of the economic analysis of the trial data and contributed to
and cost-effectiveness of support programmes posteduca- the estimation of the treatment costs used in the long-term economic analyses.
tion, so that more adults with T1DM achieve glycaemic AB, SD, NW, JE, SH and MC contributed to the design of the trial. NW, JE and SH
provided clinical input into the design of the economic analyses. MC designed the
targets. The DAFNEplus NIHR programme grant, which
statistical analysis plan. EL implemented the statistical analysis plan, parts of which
will report in 2022, is developing and testing an adapted were used in the economic evaluation. HB and DW contributed to data collection
DAFNE course (based on previous research, behaviour for the economic analyses. AB and HB designed and developed previous versions
change theory and technological support) and subse- of the economic model used in these analyses. DJP wrote the first draft. All authors
approved the final draft. DJP is the guarantor.
quent structured support to improve self-management
and glucose control.36 Funding This research was funded by the UK Health Technology Assessment
Programme (project number 08/107/01). The research was sponsored by Sheffield
In conclusion, the results indicate that it would not be Teaching Hospitals NHS Foundation Trust. The clinical sites taking part were:
cost-effective to offer pumps to all adults with T1DM in the Cambridge University Hospitals NHS Foundation Trust, Dumfries and Galloway
UK, who are currently eligible to receive a structured educa- Royal Infirmary, NHS Greater Glasgow and Clyde, Harrogate and District NHS
tion course and do not have an immediate clinical need Foundation Trust, King’s College Hospital NHS Foundation Trust, NHS Lothian,
Nottingham University Hospitals NHS Trust and Sheffield Teaching Hospitals NHS
for a pump in the UK. Use of MDI+DAFNE is estimated to Foundation Trust. We also acknowledge the financial support of the Research
represent a better use of NHS resources than immediate and Development Programmes of the Department of Health for England and
commencement on a pump. Further research is required the Scottish Health and Social Care Directorates, which supported the costs of
to improve the glycaemic control of adults with T1DM in consumables and of Medtronic UK Ltd, which provided the insulin pumps for the
trial.
the UK.
Disclaimer The views and opinions expressed herein are those of the authors and
do not necessarily reflect those of the HTA, NIHR, NHS, the Department of Health or
Author affiliations
1 Medtronic UK Ltd.
School of Health and Related Research (ScHARR), University of Sheffield, Sheffield,
UK Competing interests SH reports personal fees from Sanofi Aventis and personal
2
Population Evidence and Technologies, Warwick Medical School, University of fees and other from NovoNordisk and Eli Lilly, outside the submitted work. JE
Warwick, Coventry, UK reports personal fees from Astra Zeneca, Merck Sharpe Dohme and Takeda,
3
Academic Unit of Diabetes, Endocrinology and Metabolism, Department of personal fees and non-financial support from Eli Lilly, Novonordisk and Sanofi,
outside the submitted work. NW reports receiving two days of consultancy fees
Oncology and Metabolism, School of Medicine and Biomedical Sciences, University
from Novo Nordisk on topics including alternatives to network meta-analysis,
of Sheffield, Sheffield, UK
4 unrelated to REPOSE.
Clinical Trials Research Unit, School of Health and Related Research (ScHARR),
University of Sheffield, Sheffield, UK Patient consent Obtained.
5
RTI Health Solutions, Manchester, UK Ethics approval Research Ethics Committee (REC) North West, Liverpool East.
Provenance and peer review Not commissioned; externally peer reviewed.
Acknowledgements We would like to thank Richard Jacques and Mónica
Hernández Alava for their advice on how to conduct the statistical analysis on Data sharing statement Requests for patient level data and statistical code
HbA1c and Peter Mansell for providing helpful comments on the full report. should be made to the corresponding author and will be considered by the REPOSE
trial management group who, although specific consent for data sharing was not
Collaborators Simon Heller was the chief investigator. Norman Waugh was obtained, will release data on a case-by-case basis following the principles for
the deputy chief investigator. Stephanie Amiel, Mark Evans, Fiona Green, Peter sharing patient level data as described by Smith et al (2015). The presented data do
Hammond, Alan Jaap, Brian Kennon, Robert Lindsay and Peter Mansell were site not contain any direct identifiers; we will minimise indirect identifiers and remove
principal investigators and contributed to the study design and data interpretation. free-text data to minimise the risk of identification.
Jane Baillie, Anita Beckwith, Helen Brown, Karen Callaby, Katy Davenport, Sarah
Donald, Jackie Elliott, Leila Faghahati, Sara Hartnell, Allison Housden, Kalbir Kaur Open Access This is an Open Access article distributed in accordance with the
Pabla, Nicola Croxon, Sheena Macdonald, Muna Mohammed, Vicky Steel, Katy terms of the Creative Commons Attribution (CC BY 4.0) license, which permits
Valentine, Pamela Young, Ann Boal, Patsy Clerkin, Lynn Doran, Joanne Flynn, Emma others to distribute, remix, adapt and build upon this work, for commercial use,
provided the original work is properly cited. See: http://creativecommons.org/
Gibb, Hilary Peddie, Bernie Quinn, Helen Rogers, Janice Shephard, Janet Carling,
licenses/by/4.0/
Ann Collins, Laura Dinning, Christine Hare, Joyce Lodge, Sutapa Ray, Debora Brown,
Jenny Farmer, Alison Cox, Chris Cheyette, Pratik Choudhary, Linda East, June Ellul, © Article author(s) (or their employer(s) unless otherwise stated in the text of the
Katherine Hunt, Kimberley Shaw, Ben Stothard, Lucy Diamond, Lindsay Aniello, article) 2018. All rights reserved. No commercial use is permitted unless otherwise
Debbie Anderson, Kathy Cockerell, Vida Heaney, Alison Hutchison, Nicola Zammitt, expressly granted.
BMJ Open: first published as 10.1136/bmjopen-2017-016766 on 7 April 2018. Downloaded from http://bmjopen.bmj.com/ on 28 August 2018 by guest. Protected by copyright.
1. Nathan DM, Genuth S, Lachin J, et al. The effect of intensive subcutaneous insulin infusion (CSII) in the Veneto region: efficacy,
treatment of diabetes on the development and progression of long- acceptability and quality of life. Diabet Med 2002;19:628–34.
term complications in insulin-dependent diabetes mellitus. N Engl J 26. Gunn D, Mansell P. Glycaemic control and weight 7 years after Dose
Med 1993;329:977–86. Adjustment For Normal Eating (DAFNE) structured education in Type
2. DAFNE Study Group. Training in flexible, intensive insulin 1 diabetes. Diabet Med 2012;29:807–12.
management to enable dietary freedom in people with type 1 27. Speight J, Amiel SA, Bradley C, et al. Long-term biomedical and
diabetes: dose adjustment for normal eating (DAFNE) randomised psychosocial outcomes following DAFNE (Dose Adjustment For
controlled trial. BMJ 2002;325:746. Normal Eating) structured education to promote intensive insulin
3. Mühlhauser I, Bruckner I, Berger M, et al. Evaluation of an intensified therapy in adults with sub-optimally controlled Type 1 diabetes.
insulin treatment and teaching programme as routine management Diabetes Res Clin Pract 2010;89:22–9.
of type 1 (insulin-dependent) diabetes. The Bucharest-Düsseldorf 28. Curtis L. Unit Costs of Health & Social Care 2014. 2014. Online
Study. Diabetologia 1987;30:681–90. Source http://www.pssru.ac.uk/project-pages/unit-costs/2014/index.
4. National Institute for Health and Care Excellence. Type 1 diabetes php (accessed 30 Jun 2015).
in adults: diagnosis and management. Clinical guideline NG17. 29. British Medical Association and Royal Pharmaceutical Society of
2015. http://wwwniceorguk/guidance/ng17/evidence/full- Great Britain. British National Formulary August 2015. 2015 https://
guideline-435400241 (accessed 21 Jan 2016). www.medicinescomplete.com (accessed 9 Sep 2015).
5. National Institute for Health and Care Excellence. Continuous 30. Department of Health. NHS reference costs 2013 to 2014. 2014
subcutaneous insulin infusion for the treatment of diabetes mellitus: https://www.gov.uk/government/publications/nhs-reference-costs-
NICE technology appraisal guidance [TA 151]. 2008. https://www. 2013-to-2014 (accessed 17 Jul 2015).
nice.org.uk/guidance/ta151 (accessed 21 Jan 2016). 31. Health & Social Care Information Centre. Prescription Cost Analysis
6. Hospital and Social Care Information Centre. National Insulin Pump England 2014. 2015 http://www.hscic.gov.uk/catalogue/PUB17274
Audit Report, 2013-15. England, 2016. 32. R Core Team. R: A language and environment for statistical
7. Pickup J. Insulin pumps. Int J Clin Pract 2011;65:16–19. computing. R Foundation for Statistical Computing: Vienna, Austria,
2015. https://www.R-project.org/
8. Pickup JC. Are insulin pumps underutilized in type 1 diabetes? Yes.
33. Brown MM, Brown GC, Sharma S, et al. Quality of life with visual
Diabetes Care 2006;29:1449–52.
acuity loss from diabetic retinopathy and age-related macular
9. Cummins E, Royle P, Snaith A, et al. Clinical effectiveness and
degeneration. Arch Ophthalmol 2002;120:481–4.
cost-effectiveness of continuous subcutaneous insulin infusion
34. Roze S, Smith-Palmer J, Valentine W, et al. Cost-effectiveness of
for diabetes: systematic review and economic evaluation. Health
continuous subcutaneous insulin infusion versus multiple daily
Technol Assess 2010;14:1.
injections of insulin in Type 1 diabetes: a systematic review. Diabet
10. Colquitt JL, Green C, Sidhu MK, et al. Clinical and cost-effectiveness
Med 2015;32:1415–24.
of continuous subcutaneous insulin infusion for diabetes. Health
35. Weissberg-Benchell J, Antisdel-Lomaglio J, Seshadri R. Insulin pump
Technol Assess 2004;8:1–171.
therapy: a meta-analysis. Diabetes Care 2003;26:1079–87.
11. Leotta SA, Altomare M, Carletti S, et al. Intensive diabetes
36. National Institute for Health Research. Programme Grants for Applied
management in type 1 diabetic patients in poor glycemic control
Research. Active, completed and closed programme grants. Online
treated with insulin pump therapy or multiple daily injections:
Source https://www.nihr.ac.uk/research-and-impact/documents/
Two years follow-up. 4th International Conference on Advanced Funded%20studies/PGfAR%20funded%20projects%20June%
Technologies and Treatments for Diabetes, 2011. 202017.xlsx (accessed 15 Nov 2017).
12. White D, Waugh N, Elliott J, et al. The Relative Effectiveness of 37. DAFNE. DAFNE Fact Sheet Six. 2012. Online Source http://www.
Pumps Over MDI and Structured Education (REPOSE): study dafne.uk.com/uploads/135/documents/06_factsheetsix_12pt_18_06_
protocol for a cluster randomised controlled trial. BMJ Open 12.pdf (accessed 1 Jul 2015).
2014;4:e006204. 38. Department of Health. NHS Reference Costs 2010-11. London:
13. National Institute for Health and Care Excellence. Guide to the Department of Health, 2011.
methods of technology appraisal 2013. 2013. Online Source http:// 39. British Medical Association and Royal Pharmaceutical Society of
www.nice.org.uk/article/pmg9/resources/non-guidance-guide-to-the- Great Britain. British National Formulary. 63rd edn. London: BMA and
methods-of-technology-appraisal-2013-pdf (accessed 13 Jul 2015). RPS, 2012.
14. Heller S, Lawton J, Amiel S, et al. Improving management of type 1 40. McEwan P, Poole CD, Tetlow T, et al. Evaluation of the cost-
diabetes in the UK: the Dose Adjustment For Normal Eating (DAFNE) effectiveness of insulin glargine versus NPH insulin for the treatment
programme as a research test-bed. A mixed-method analysis of the of type 1 diabetes in the UK. Curr Med Res Opin 2007;23:S7–S19.
barriers to and facilitators of successful diabetes self-management, 41. Currie CJ, Poole CD, Woehl A, et al. The financial costs of healthcare
a health economic analysis, a cluster randomised controlled trial of treatment for people with Type 1 or Type 2 diabetes in the UK with
different models of delivery of an educational intervention and the particular reference to differing severity of peripheral neuropathy.
potential of insulin pumps and additional educator input to improve Diabet Med 2007;24:187–94.
outcomes. Programme Grants for Applied Research 2014;2:1–188. 42. National Institute for Health and Care Excellence. Ranibizumab
15. Thokala P, Kruger J, Brennan A, et al. Assessing the cost- for treating diabetic macular oedema (rapid review of technology
effectiveness of type 1 diabetes interventions: the Sheffield type 1 appraisal guidance 237). NICE technology appraisal guidance 274.
diabetes policy model. Diabet Med 2014;31:477–86. 2013. Onlince Source https://www.nice.org.uk/guidance/ta274/
16. SIMUL8 corporation. SIMUL8 professional 2010. Boston, 2010. resources/ranibizumab-for-treating-diabetic-macular-oedema-rapid-
17. Groot Koerkamp B, Stijnen T, Weinstein MC, et al. The combined review-of-technology-appraisal-guidance-237-82600612458181
analysis of uncertainty and patient heterogeneity in medical decision 43. British Medical Association and Royal Pharmaceutical Society of
models. Med Decis Making 2011;31:650–61. Great Britain. British National Formulary August 2016. 2016 https://
18. REPOSE Study Group. Relative effectiveness of insulin pump www.medicinescomplete.com
treatment over multiple daily injections and structured education 44. Clarke P, Gray A, Legood R, et al. The impact of diabetes-related
during flexible intensive insulin treatment for type 1 diabetes: cluster complications on healthcare costs: results from the United Kingdom
randomised trial (REPOSE). BMJ 2017;356:j1285. Prospective Diabetes Study (UKPDS Study No. 65). Diabet Med
19. Statacorp. Stata Statistical Software. Release 13.1. College Station, 2003;20:442–50.
Texas Stata Press, 2013. 45. Peasgood T, Brennan A, Mansell P, et al. The impact of diabetes-
20. Kaplan EL, Meier P. Nonparametric Estimation from Incomplete related complications on preference-based measures of health-
Observations. J Am Stat Assoc 1958;53:457–81. related quality of life in adults with Type I diabetes. Med Decis
21. Collett D. Survival Data in Medical Research. 2nd ed. Boca Ration Making. In Press. 2016;36:1020–33.
(FL: Chapman & Hall / CRC, 2003. 46. Coffey JT, Brandle M, Zhou H, et al. Valuing health-related quality of
22. Cribari-Neto F, Zeileis A. Beta Regression in R. J Stat Softw life in diabetes. Diabetes Care 2002;25:2238–43.
2010;34:1–24. 47. Alva M, Gray A, Mihaylova B, et al. The effect of diabetes
23. Orr CJ, Hopman W, Yen JL, et al. Long-term efficacy of insulin pump complications on health-related quality of life: the importance of
therapy on glycemic control in adults with type 1 diabetes mellitus. longitudinal data to address patient heterogeneity. Health Econ
Diabetes Technol Ther 2015;17:49–54. 2014;23:487–500.
24. Carlsson BM, Attvall S, Clements M, et al. Insulin pump-long-term 48. Clarke P, Gray A, Holman R. Estimating utility values for health states
effects on glycemic control: an observational study at 10 diabetes of type 2 diabetic patients using the EQ-5D (UKPDS 62). Med Decis
clinics in Sweden. Diabetes Technol Ther 2013;15:302–7. Making 2002;22:340–9.